That link is a really interesting read. In particular, this quote struck me:
Intel's CPUs use techniques such as having 5 Floating Point
Units designed by separate teams, having them all on-board the CPU, and
asking every unit to perform the same calculation. As the units
are all by different teams, with different designs, they have different
advantages, given different ranges of inputs. The fastest to complete the
calculation is chosen, each time, and all other partial answers abandoned.
The resources wasted, all in the name of "speed"...
I've never heard this before. A casual search turned up nothing, how do you know this?